The Week of AI Agents: Hermes vs. OpenClaw, Linux Security Vulnerabilities and Open-Source Tools in Focus
Thursday, May 14, 2026
Hello, this weekly newsletter guides you through the most important new videos from a curated selection of AI and Coding YouTube channels. Each video gets a compact summary, plus a daily overview of the dominant topics. If interested, simply click the link under the summary.
This week is all about AI agents and their applications. The comparisons between Hermes and OpenClaw, which were thoroughly tested and evaluated in several videos, are particularly in the spotlight. Alex Finn conducted extensive tests in which both agents competed against each other in various scenarios. It became clear that Hermes excels in some areas like user-friendliness and self-improvement, while OpenClaw was praised for its stability and consistency. The discussions about the strengths and weaknesses of both agents dominated the week and showed how rapidly the landscape of AI tools is evolving.
Another central topic was the security vulnerability in the Linux kernel discovered by an AI tool. Fireship reported in detail on the “copy fail” vulnerability, which has existed since 2017 and can be exploited through a Python script. The emphasis was on the role of AI in discovering such security vulnerabilities and the need to update systems regularly. The supply chain attack on open-source packages in the npm ecosystem was also discussed, with tools like PNPM and Sentry proposed as solutions.
The week was characterized by a strong dominance of open-source tools and models. From creating AI agents with Hermes to using OpenCode to employing free models like Qwen 3.6 and JML4 – the open-source community demonstrated its strength and diversity. Particularly noteworthy is the video by Leon van Zyl, who demonstrated how to use the open-source tool “Honeyfree” to autonomously plan and implement software projects. These tools not only offer cost-free alternatives but also provide high customizability and privacy.
A special highlight was Alex Finn’s video in which he set up a Hermes agent on a local model on an Nvidia DGX Spark. The demonstration of use cases such as daily reports on AI stocks and the creation of to-do list apps showcased the potential of local models. The discussion about the strategic alliance between Anthropic and Elon Musk’s XAI was also an exciting topic that changed the dynamics in AI competition. This week once again showed how rapidly technology is advancing and how important it is to stay up to date.
Niklas Steenfatt
No new videos in this period.
Fireship (3 new videos)
- Every operating system concept in one video…
7.5.2026, 17:32:34The video explains in detail how an operating system works from the moment the power button is pressed until shutdown. It begins with the bootloader, which loads the operating system, then moves on to privilege rings, which separate kernel and application permissions. Virtual memory is described as a system that allows multiple applications to run in parallel without interfering with each other. The kernel builds the file system, loads device drivers, and enables interrupts, which allow the system to respond to input. The kernel then starts the first process (PID1), which is the ancestor of all other processes. System calls enable applications to communicate with the kernel, and the scheduler manages CPU time for the many processes. Threads allow applications to execute multiple tasks simultaneously, and inter-process communication (IPC) enables different processes to communicate safely. Finally, the shutdown process is described, where all processes are terminated and the system shuts down safely.
The video covers operating systems and their components in general without mentioning specific tools or vendors, and is more suitable for intermediate or advanced viewers.
- 732 bytes of Python just borked every Linux machine on earth…
4.5.2026, 18:40:40The video covers a critical security vulnerability in the Linux kernel, referred to as “copy fail” (CVE-2023-31431), which has existed since 2017 and was discovered by an AI tool. The vulnerability allows a local user to gain root access by writing four bytes to the page cache of a read-only file. All Linux distributions updated after 2017 are affected. The vulnerability was exploited through a Python script that uses the ONC ESN protocol and AF_AGL interface. Although the vulnerability is not remotely exploitable, it is strongly recommended to update systems. The video also mentions the role of AI in discovering security vulnerabilities and promotes Code Rabbit, an AI tool for improving code quality.
The video explicitly covers AI tools like the AI agent tool used by Theori and Code Rabbit, and is intended for intermediate to advanced users.
- A single PR just hijacked the NPM registry…
14.5.2026, 17:39:11The video reports on a severe supply chain attack on open-source packages in the npm ecosystem, where over 100 packages with a combined download rate of more than 50 million times per week were compromised. The attack exploited a vulnerability in the Tanstack project’s release process by having an attacker create a pull request in a fork of the repository, which triggered the CI/CD workflow. By using the `pull_request_target` option, the attacker was able to inject malicious code into the CI server’s shared cache, which was later triggered by a legitimate merge. The malware used stolen npm publish tokens to publish further compromised packages and spread to other systems, including infecting developer tools like GitHub and VS Code. The malware even included a “dead man switch” that deleted root directories of infected machines upon detection.
To protect against such attacks, the video recommends using PNPM version 1 or higher, which offers features like “Minimum Release Age”, “Block Exotic Subdependencies”, and “Approved Builds” to block malicious packages. Additionally, Sentry is presented as a tool for error tracking in production that works with an AI-powered agent to automatically investigate and fix issues.
The video covers specific tools such as PNPM, npm, GitHub Actions, and Sentry and is aimed at intermediate to advanced users.
Alex Finn (8 new videos)
- LIVE: The greatest Claude Code workflow ever
13.5.2026, 20:12:33The video showcases a detailed, advanced workflow for Claude Code that integrates various tools such as Slack, Linear, GitHub, and Claude Code. The host explains how these tools work together to boost productivity, track changes, and organize development. The workflow includes creating tasks and projects in Linear, linking them with GitHub for branch management, and using Claude Code to automate and manage these processes. The host emphasizes the benefits of this workflow, including increased speed, better error prevention, and improved traceability.
Additionally, the host discusses personal experiences and challenges, such as dealing with difficult times and the importance of perseverance. He also shares his thoughts on using AI tools like Claude Code and Codex, as well as their differences and use cases.
The video is intended for advanced users who already have experience with Claude Code and similar tools and want to optimize their workflows. Specific tools such as Claude, OpenAI, and Linear are discussed.
- Hermes Agent powered by local models on the DGX Spark is basically magic
13.5.2026, 13:30:07The video demonstrates how to set up a Hermes Agent on a local model on an Nvidia DGX Spark to create a 24/7 available AI employee. The process includes setting up the DGX Spark in headless mode, installing a local model (Qwen 3.6 27B), and integrating the model into the Hermes Agent. The creator demonstrates three use cases: a daily report on AI stocks for beginners, repurposing YouTube video content for advanced users, and vibe coding a to-do list app for advanced users. The focus is on the benefits of local models, including cost-free operation (apart from electricity consumption), privacy, customizability, and educational value.
Closing note: The video explicitly addresses Nvidia DGX Spark, Hermes Agent, Qwen 3.6 27B, and Tail Scale and is intended for intermediate to advanced users.
- LIVE: Talking AI news (no Hermes use cases ignore the thumbnail)
11.5.2026, 20:11:47The video is a live stream session that primarily revolves around discussing Hermes Agent and its use cases. The host, Alex Finn, begins with an introduction to Hermes Agent and emphasizes the importance of use cases for utilizing the technology. He mentions that Hermes has released a new website with hundreds of use cases that he and the viewers will go through and test.
However, a large portion of the video is taken up by various tangents and discussions, including:
1. **Investment Corner**: Alex discusses current investment opportunities in the AI industry, particularly in companies like Nvidia, Micron, TSMC, and Tesla. He emphasizes the importance of current AI developments and the need to invest in the right companies.
2. **Personal Stories and Anecdotes**: Alex shares personal stories and anecdotes that are often humorous and entertaining but not directly related to the main topic of the video.
3. **Chat Interaction**: A large part of the video consists of interaction with the live chat, with Alex answering questions, making comments, and engaging with viewers.
4. **Use Cases for Hermes Agent**: Toward the end of the video, Alex begins to discuss some of the use cases from the new Hermes website. He mentions use cases such as creating research reports, managing tasks, and automating processes.
5. **Announcements and Updates**: Alex provides updates on his own projects and announcements, such as the launch of a second YouTube channel and attending a Baby Keem concert.
**Closing note**: The video explicitly addresses open-source AI models and tools such as Hermes Agent and OpenClaw. It is intended for intermediate to advanced users who already have some understanding of AI and its applications.
- Hermes Agent is blowing me away…
9.5.2026, 20:54:26The video compares the AI agents Hermes and OpenClaw and recommends Hermes due to its reliability, self-improvement, and user-friendliness. The author describes the benefits of Hermes, including regular, thematic updates, self-improvement capabilities through usage, and a strong emphasis on experimentation and local models. Installing Hermes is described as straightforward, with options for various models and communication services, with Telegram and Opus recommended. The author showcases three use cases: one for beginners that discovers new AI tools daily, one for advanced users that performs daily proactive check-ins, and one for experts that creates AI-generated videos. At the end, the author emphasizes the importance of brain dumping and reverse prompting to use the AI agent personally and effectively.
The video explicitly addresses the AI tools Hermes Agent and OpenClaw and is intended for intermediate and advanced users.
- LIVE: Anthropic and Elon just teamed up to take down OpenAI
6.5.2026, 20:12:34The video covers the strategic alliance between Anthropic and Elon Musk’s XAI (X.AI), characterized by a major computing power deal. Anthropic gains access to SpaceX’s Colossus-1 cluster, which will significantly enhance their ability to develop and train AI models. This partnership marks a turning point in competition with OpenAI, which has taken a dominant position in recent months with Codeex. During this time, Anthropic has struggled with declining limits and less powerful models, which the new alliance aims to address. Elon Musk, who was previously critical of Anthropic, is now providing massive computing resources, changing the dynamics of AI competition. The video also discusses Elon Musk’s long-term strategies, which may focus on larger goals such as autonomous vehicles, space travel, and robotics rather than staying in the AI chatbot competition. The alliance could lead to a new era of innovation and improvement in AI tools, from which consumers will benefit. The video emphasizes the importance of using both leading AI tools, Claude Code and Codeex, to benefit from their respective strengths.
**AI Tools/Models/Providers:** Anthropic, OpenAI, Elon Musk (X.AI), Claude, Codeex, Grok, Gemini, Open-Source
**Target Audience:** Intermediate - Hermes Agent might have just killed OpenClaw
5.5.2026, 21:11:59The video presents Hermes Agent as a more reliable alternative to OpenClaw and covers seven new features that improve productivity and user-friendliness. These include:
1. **Kanban Board**: Enables multitasking through parallel processing of multiple task threads. A manager agent populates tasks with details and moves them through various statuses (Triage, To-Do, Ready, In Progress, Blocked, Done).
2. **Slash Goals**: A high-level mission function that assigns the agent long-term tasks that can be worked on over an extended period. The quality of the prompt is crucial for good results.
3. **Profiles (Multi-Agents)**: Allows the creation of multiple agents with their own memories and capabilities to optimize performance and prevent overload.
4. **Model Catalog**: Simplifies switching and assigning models to specific tasks, improving cost control and efficiency.
5. **Compression**: By adjusting the compression threshold to 0.5, less drastic compressions are performed, improving memory retention.
6. **Curator Feature**: Automatically prunes rarely used skills every seven days to reduce bloat and maintain performance.
The video criticizes OpenClaw for frequent updates that lead to instability and performance issues, while highlighting Hermes’s targeted, reliable updates. It’s recommended to use Hermes’s new features to boost productivity.
**Closing note**: The video explicitly addresses Hermes Agent and OpenClaw and is intended for intermediate to advanced users.
- LIVE: Is Hermes better than OpenClaw? FINALE!!!
4.5.2026, 21:53:53The YouTube video shows a live stream where the host tests various AI agents (OpenClaw and Hermes) in a competition called “Agent Olympics.” The stream is unusually long (3.5 hours) and is divided into various segments ranging from technical tests to personal discussions to spontaneous decisions.
**Content Summary:**
1. **Agent Olympics:**
– The host tests four combinations of AI agents (OpenClaw and Hermes with different backend models such as ChatGPT and Opus) in five different tasks.
– The tasks include creating infographics, animated music videos, and other complex assignments.
– Results are evaluated live, with OpenClaw with Opus emerging as the winner at the end.2. **Technical Discussions:**
– There are extensive discussions about the stability and reliability of the various AI agents, particularly Hermes, which is criticized due to “compaction” errors (loss of work states).
– OpenClaw is praised for its consistency and user-friendliness.3. **Personal Topics:**
– The host discusses his sleep problems and experiments with various solutions such as kiwis and magnesium.
– There are discussions about work methods, including the use of treadmills and standing desks, with the host expressing his preferences and dislikes.4. **Community Interaction:**
– Viewers are actively engaged in the chat, asking questions and providing feedback.
– The host spontaneously decides to create a second YouTube channel called “Alex Finn Labs,” leading to an entertaining interaction with a viewer who has already reserved the desired channel name.5. **Announcements and Future Plans:**
– The host announces plans to publish more videos about Hermes and multi-agent setups in the future.
– There is discussion about whether live streams should take place at later times to reach a broader audience.**Closing note:**
The video explicitly addresses the AI tools OpenClaw, Hermes, ChatGPT, and Opus. It is intended for intermediate and advanced users as it covers technical details and advanced applications of AI agents. - LIVE: OpenClaw vs Hermes Agent: The ultimate showdown
1.5.2026, 20:48:54The YouTube video shows a live stream where various AI agents are tested in direct comparison. The main participants are OpenClaw and Hermes, each running with the models ChatGPT and Opus. The stream is divided into several tests that evaluate the capabilities of the agents in various task areas.
1. **Test 1: Real-Time Stock Dashboard**
– **OpenClaw with ChatGPT**: Fastest completion, but with an unattractive user interface (UI). Functionality was rated as solid.
– **Hermes with ChatGPT**: Slower and crashed the computer, resulting in a poor rating.
– **OpenClaw with Opus**: Slower than the ChatGPT version, but with a slightly better UI and additional features such as TradingView integration.
– **Hermes with Opus**: Best UI and functionality, but slower than OpenClaw with ChatGPT.2. **Test 2: Game Development**
– **OpenClaw with ChatGPT**: Fast, but unplayable game.
– **OpenClaw with Opus**: Playable, but not particularly entertaining.
– **Hermes with ChatGPT**: Unplayable and poor graphics.
– **Hermes with Opus**: Best graphics and playability, rated as actually entertaining.3. **Test 3: Website Recreation (Apple.com)**
– **Hermes with Opus**: First to finish, but below-average accuracy.
– **OpenClaw with Opus**: Better than Hermes with Opus, but not perfect.
– **OpenClaw with ChatGPT**: Most accurate, nearly perfect.
– **Hermes with ChatGPT**: Perfect recreation through screenshots, but ethically questionable.The stream ends with Hermes with Opus in the lead, followed by OpenClaw with Opus and OpenClaw with ChatGPT. Hermes with ChatGPT lags significantly behind. The remaining tests will continue in the next live stream.
**Closing note**: The video explicitly addresses the AI models Claude (Opus), OpenAI (ChatGPT), and specific tools such as OpenClaw and Hermes. It is intended for intermediate and advanced users interested in the performance and comparison of AI agents.
Leon van Zyl (9 new videos)
- Claude Code Agent View: Parallel Agents Are Here
14.5.2026, 10:51:58The video reports on a complex supply chain attack targeting open-source packages in the npm ecosystem, affecting over 100 packages with more than 50 million weekly downloads. The attack exploited a vulnerability in Tanstack’s release process, where an attacker created a pull request in a fork of the repository, triggering the CI/CD workflow. By using the “pull request target” option, the attacker was able to inject malicious files into the CI server’s shared cache, which later stole a valid npm publish token and published compromised package versions. The malware spread further by searching for additional npm publish tokens and infecting more packages, including those from Mistral AI, UiPath, and Open Search. The malware deeply integrated into developer environments and even employed a “dead man switch” that deleted the root directory of infected systems upon detection.
To defend against such attacks, the video recommends using PNPM 1 or higher, which offers features like “minimum release age,” “block exotic subdependencies,” and “approved builds” to prevent installation of malicious packages. Additionally, Sentry is introduced as a tool for error management in production, working with an AI-powered agency called Seir to automatically investigate and fix issues.
The video addresses specific tools such as PNPM, Sentry, and Seir Agent and is aimed at intermediate to advanced users.
- Codex CLI Tutorial: Build an AI Image Studio from Scratch
11.5.2026, 11:17:21This video walks through building an AI image studio for creating YouTube thumbnails, posters, banners, and other graphics step by step. The process starts by setting up a Next.js project using the Codex CLI tool, based on GPT-5.5 and high reasoning levels. A database is configured with Docker and Postgres, and necessary tables for user authentication are migrated.
The focus is on creating a user-friendly interface that allows users to upload reference images, write prompts, and generate images. The developer leverages Codex’s capabilities to design and test the user interface while adhering to a predefined design system. After designing the interface, the actual functionality is implemented using an OpenAI API key for the GPT-4 image model. The developer demonstrates how to upload reference images and assets, and how to generate thumbnails combining these elements.
At the end, the user interface is adjusted to ensure image generation only occurs through the dashboard, not from the homepage. The homepage is redesigned with a generated image and marketing text. The video emphasizes the efficiency and time savings achieved through using Codex and OpenAI, while noting the token limitations of the ChatGPT Plus plan.
The video explicitly covers OpenAI (GPT-5.5, GPT-4 Image Model) and Codex. It’s geared toward intermediate and advanced users since it addresses advanced concepts like Docker, Postgres, Next.js, and API integration.
- Create Custom OpenCode Agents #Shorts #OpenCode #AICoding
10.5.2026, 07:00:16The video shows how to create custom agents in OpenCode. By default, there are two agents: “build” and “plan.” To create a custom agent, you run the command `opencode agent create` in the terminal. Next, you provide a description of the agent—for example, that an agent named John responds only in emojis. After creation, you can use the spacebar to define which functions or tools the agent can access. You also select the agent mode: either for primary and subordinate roles or only as a subordinate agent. The video demonstrates creating a subordinate agent named John and shows how the main agent delegates a task to John. You can track the subordinate agent’s work and view its reasoning and outputs. Finally, it emphasizes that this is just a demonstration and that in practice you could use more specific system prompts and tool access for subordinate agents.
The video covers OpenCode and is suitable for intermediate users.
- OpenCode’s Best Hidden Feature #Shorts #OpenCode #AICoding
9.5.2026, 07:00:19The video explains how to configure different models for different modes in OpenCode. For example, you can use a powerful model like GPT-5 for planning mode and a fast, cost-effective model like Big Pickle for implementation mode (Bold Mode). Alternatively, you can use GPT-5.5 for planning but reduce effort by selecting the “/variants” command with the “low effort” option. This way, planning is done by an intelligent model while implementation is handled by a less powerful but faster model.
The video covers open-source models and is more suitable for intermediate users.
- I Turned Hermes Agent Into a Coding Agent
8.5.2026, 11:02:33The video demonstrates how to use the Hermes Agent as a coding agent to create a web app and deploy it online. The process includes setting up the Hermes Agent on a VPS, integrating it with Telegram for communication, installing the Vercel CLI tool for deployment, and configuring the necessary skills for the agent. The creator tests whether the agent can create a personal portfolio page by scraping information from the creator’s YouTube channel and building an appealing frontend. The agent successfully creates the app, deploys it on Vercel, and provides a public URL that opens the app in a browser. The video also shows that the agent can make changes to the app and deploy them automatically.
The creator concludes that Hermes is suitable as a coding agent for simple tasks and quick dashboards but not for complex software projects. The video explicitly covers Hermes Agent, OpenAI Codex, GPT 5.5, Vercel, and Telegram. It’s aimed at intermediate and advanced users.
- This free OpenCode trick saves thousands #opencode #aitools #hacks
7.5.2026, 13:38:39The video explains how to use free AI models for code generation in OpenCode. To do this, you first run the “connect” command and search for “OpenCode Zen” under providers. Next, you generate an API key through a provided URL, which is free. After entering the API key, you get a list of supported models, including Big Pickle, HY3, Minimax M2.5, and Neurotron 3 Super from Nvidia. These models are powerful and completely free.
The video covers OpenCode and OpenCode Zen and is aimed at intermediate users.
- I Built a Coding Agent That Runs Locally for Free
6.5.2026, 12:25:00The video introduces the open-source tool “Honeyfree,” which enables autonomous planning and implementation of software projects. You describe to the tool what you want to build, and it plans the features, adds them to a Kanban board, and implements them automatically. The tool supports various models like Alum Studio and Ollama and can break down complex tasks into smaller features. The creator demonstrates creating a simple to-do app and shows how new features can be added and implemented. The video emphasizes that this is now possible with free models, which wasn’t the case a few months ago. It also explains how to download models like Qwen 3.6 or JML4 and use them in Alum Studio or Llama Studio. The creator recommends increasing the context window length of models to at least 64,000 tokens for better performance. The video also shows how to install and set up Local Forge to create and manage projects. It stresses that while free models are good at writing code, they benefit from more detailed instructions for better results. The creator recommends using a paid model like Claude for planning features while using free models for actual implementation. The video ends with an invitation to sign up for a masterclass course teaching how to build applications with coding agents.
The video covers open-source models like Qwen 3.6 and JML4, as well as tools like Alum Studio, Llama Studio, and Local Forge. It’s suited for intermediate and advanced users who already have experience using AI models and developing software.
- OpenCode Tutorial for Beginners: Setup, Agents, Skills & MCP
5.5.2026, 12:33:17This video is a tutorial showing how to create a Next.js application with OpenCode, an open-source AI tool. The process starts with installing and setting up OpenCode, including connecting to various AI models and providers, both free and paid. The tutorial demonstrates how to add agent skills like frontend design and Next.js skills to improve the quality of generated code. It also shows how to use memory files and design systems to increase the agent’s consistency and efficiency. The tutorial continues with creating an application that allows users to input a rough idea of their app and receive a detailed project plan. The agent uses subagents to execute tasks in parallel and protect the main context. At the end, the application is tested and improved, with the agent completely redesigning the UI and running automated tests. The video is suited for intermediate and advanced users interested in AI-powered coding tools.
AI Tools/Models/Providers: OpenCode (open-source), OpenAI, Anthropic, Gemini, OpenRouter, BigPicko, HY3, Minimax, Nvidia, Vercel, Cintra AI.
- I Built a Full App Using Only Cursor AI
4.5.2026, 11:01:35In this video, an AI-powered YouTube summarizer is developed using the Cursor tool. The process begins with creating a user interface that takes a YouTube URL and provides a video summary. Requirements include a brief summary (TLDR), five to eight key points, a “Watch these moments” section with timestamps and descriptions, and the video’s original link.
The creator uses Cursor and chooses the Composer 2 model to scaffold the project. Next.js is installed along with the Shad cn library for the user interface. With Cursor’s agents, a basic user interface is created meeting the requirements. Functionality is then added to retrieve a YouTube video’s transcript using the YouTube Transcript API.
For AI-powered summarization, the Cursor AI SDK is used to return structured data. The creator opts for the “anthropic/claude-2” model from OpenRouter and integrates the API key into an .env file. The agent then generates the summary, including the TLDR, key points, and recommended moments from the transcript.
The video explicitly covers the tools Cursor, Composer 2, Next.js, Shad cn, YouTube Transcript API, AI SDK, and OpenRouter. It’s more suited for intermediate and advanced users.
Automatically generated from the latest YouTube videos of the curated channel selection. For feedback, suggestions, or to unsubscribe: simply reply to this email.